Intelligent Detection and Real-time Monitoring of Engine Oil Aeration Using a Machine Learning Model

Abstract

This research work develops a machine learning model for detecting and real-time monitoring engine oil aeration in an internal combustion engine using only single high-speed oil pressure sensor. The presented method uses a five level cascading discrete wavelet transform with Daubechies 4 tap wavelet and an associated variance metric to identify features related to oil aeration from a set of recorded oil pressure traces. A Gaussian process regression model is then used to correlate the identified features to measured oil aeration and the presented approach is successfully able to predict engine oil aeration to an uncertainty of under ±0.02 from the measured oil aeration values. The sensitivity of this method to varying sampling frequencies is also tested and the method is found to be successful over a wide range of sampling frequencies. This method of predicting measured oil aeration using a single high-speed oil pressure sensor has the benefit of monitoring engine oil aeration without the need for direct measurement.